Balancing Auto-coding and Manual Coding in MedDRA: Risks and Benefits

Introduction to Auto-coding and Manual Coding

Adverse event reporting in clinical trials depends heavily on MedDRA coding. Coders and pharmacovigilance staff transform investigator-reported verbatim terms into standardized Lowest Level Terms (LLTs) and Preferred Terms (PTs). Two primary approaches exist: auto-coding and manual coding. Both methods are widely used, and most sponsors employ a hybrid approach to balance efficiency and accuracy.

Auto-coding refers to the use of software algorithms that automatically map verbatim terms to MedDRA LLTs and PTs. This process improves speed and consistency but carries risks of misclassification. Manual coding, by contrast, requires trained coders to review verbatim terms and select the most accurate PT, ensuring clinical accuracy but requiring more time and resources.

Regulatory authorities, including the FDA, EMA, MHRA, and CDSCO, accept either method, provided coding is accurate, consistent, and traceable. Inspections often focus on whether sponsors have controls to minimize auto-coding errors and whether manual coding is performed with adequate SOPs and training.

Benefits of Auto-coding

Auto-coding offers several advantages:

• Speed: Automated mapping allows high-volume processing of adverse events, especially in late-phase or large-scale trials.
• Consistency: Ensures identical verbatim terms are mapped to the same PT, reducing variability between coders.
• Efficiency: Minimizes manual workload for straightforward terms, freeing safety teams for complex coding tasks.
• Scalability: Particularly useful in global pharmacovigilance databases handling thousands of SAE and AE reports daily.

For example, common terms such as “headache,” “nausea,” or “fever” can be reliably auto-coded to their respective PTs with little risk of error. In such scenarios, auto-coding significantly improves throughput without compromising accuracy.

Risks of Auto-coding

Despite its advantages, auto-coding presents risks:

• Misclassification: Verbatim terms that are ambiguous or unusual may be incorrectly coded.
• Lack of clinical context: Algorithms may select PTs that miss subtle nuances in the investigator’s description.
• False confidence: Users may rely too heavily on automated systems without appropriate review.
• Regulatory findings: Incorrect PT assignments discovered during inspections can be classified as major findings.

A common example is the investigator term “fainting.” An auto-coding algorithm may map this to “Loss of consciousness,” while the clinically correct PT should be “Syncope.” Without manual review, the coding would be inaccurate and potentially misleading in safety analyses.

Benefits of Manual Coding

Manual coding by trained professionals provides several advantages over automation:

• Clinical judgment: Coders apply medical knowledge to interpret ambiguous or complex terms.
• Accuracy: Reduces risk of misclassification by considering clinical context and study-specific nuances.
• Flexibility: Allows handling of rare events not typically recognized by auto-coding algorithms.
• Audit readiness: Demonstrates human oversight in coding processes, which regulators value during inspections.

For example, the term “liver swelling” might not have a straightforward auto-coded PT. A trained coder would correctly assign “Hepatomegaly,” ensuring data accuracy.

Limitations of Manual Coding

Manual coding, however, has its drawbacks:

• Time-consuming: Large datasets with thousands of AEs require significant manpower.
• Inter-coder variability: Different coders may select different PTs for the same term without clear conventions.
• Resource intensive: Requires continuous training and staffing.

For global trials, where thousands of SAE reports may be received monthly, manual-only coding can strain resources and delay reporting timelines.

Hybrid Approach: Best of Both Worlds

Most sponsors adopt a hybrid approach that combines the efficiency of auto-coding with the accuracy of manual coding:

• Auto-coding: Used for common, low-risk terms like “headache” or “nausea.”
• Manual coding: Applied to ambiguous, rare, or complex terms requiring clinical interpretation.
• Quality checks: Safety departments conduct routine audits to identify and correct auto-coding errors.

This balanced method ensures that large volumes of routine data are processed efficiently, while complex cases receive the clinical oversight they require. Many sponsors implement a “70/30 split,” where 70% of coding is auto-coded and 30% is manually reviewed.

Regulatory Expectations and Inspections

Regulators expect sponsors to demonstrate oversight in both auto-coding and manual coding processes. Common inspection findings include:

• Over-reliance on auto-coding without adequate review.
• Failure to document manual coding decisions.
• Lack of SOPs governing auto-coding thresholds and exceptions.
• Inconsistent coding across studies due to inter-coder variability.

To mitigate these risks, sponsors should maintain detailed SOPs, perform reconciliation checks, and train coders on both methods. Reference registries like the Japan Registry of Clinical Trials highlight the importance of coding accuracy in safety reporting worldwide.

Key Takeaways

The choice between auto-coding and manual coding in MedDRA is not binary. Clinical trial sponsors should:

• Leverage auto-coding for routine terms to improve speed and consistency.
• Apply manual coding for complex, ambiguous, or high-risk terms.
• Adopt hybrid models with built-in quality controls.
• Ensure SOPs and conventions are updated with each MedDRA release.
• Maintain inspection readiness by documenting coding workflows and training logs.

By balancing the benefits and risks of both methods, sponsors can ensure that safety data is coded efficiently, accurately, and in line with global regulatory expectations.

Making Sense of the Regulatory Framework for Post-Market Vaccine Safety

What the Framework Covers: From Law and Guidance to Day-to-Day Controls

“Regulatory framework” sounds abstract until you are the person who must file a 15-day serious unexpected case, update a Risk Management Plan (RMP), and walk an inspector through your audit trail—all in the same week. For vaccines, the framework spans law (e.g., national medicine acts; 21 CFR in the U.S.), regional guidance (EU Good Pharmacovigilance Practice—GVP), and global harmonization (ICH E-series for safety). These documents translate into practical obligations: how to collect and submit Individual Case Safety Reports (ICSRs) using ICH E2B(R3); how to code with MedDRA and de-duplicate; how to manage signals (ICH E2E) and summarize safety/benefit-risk in periodic reports (ICH E2C(R2) PBRER/PSUR). For vaccines specifically, regulators also look for active safety and effectiveness activities that complement passive reporting—observed-versus-expected (O/E) analyses, self-controlled case series (SCCS), and post-authorization effectiveness studies that inform policy.

A credible system connects obligations to operations: a PV System Master File (PSMF) that maps processes and vendors; a validated safety database with Part 11/Annex 11 controls; ALCOA-proof documentation in the Trial Master File (TMF); and cross-functional governance (clinical, epidemiology, statistics, quality, regulatory). Quality context matters, too: reviewers often ask whether a safety pattern could reflect manufacturing or hygiene rather than biology. Keep concise statements ready—e.g., representative PDE for a residual solvent of 3 mg/day and cleaning MACO of 1.0–1.2 µg/25 cm²—alongside analytical transparency when labs inform case definitions (assay LOD 0.05 µg/mL; LOQ 0.15 µg/mL for a potency HPLC, illustrative). For SOP checklists and submission cross-walks, teams often adapt resources from PharmaRegulatory.in. For public expectations and vocabulary to mirror in filings, see the European Medicines Agency.

Expedited Reporting, Periodic Reports, and RMPs: The Heart of Compliance

Expedited case reporting is the day-to-day heartbeat of PV. Most jurisdictions require 15-calendar-day submission of serious and unexpected ICSRs from the clock-start (the first working day the Marketing Authorization Holder has minimum criteria: identifiable patient, reporter, suspect product, and adverse event). Domestic deaths may be due within 7 days in some markets (with a follow-up by Day 15). Submissions must be ICH E2B(R3)-compliant, with consistent MedDRA coding, deduplication rules, translations, and audit trails for any field edits. Periodic reporting completes the picture: PBRER/PSUR (ICH E2C(R2)) integrates cumulative safety, new signals, and benefit-risk conclusions, while Development Safety Update Reports (DSURs) may still apply in certain post-authorization studies. The RMP describes important identified and potential risks, missing information, routine/ additional pharmacovigilance, and risk-minimization measures; vaccine RMPs often include enhanced surveillance for AESIs like anaphylaxis, myocarditis, TTS, and GBS, plus effectiveness monitoring where policy depends on waning and boosters.

Every obligation should appear as a measurable control in your QMS: case-clock start/stop definitions and SLAs; coding conventions; medical review and causality procedures (WHO-UMC); and handoffs to labeling if a signal graduates to an important identified risk. When labs govern case inclusion (e.g., high-sensitivity troponin I for myocarditis), the method sheet with LOD / LOQ, calibration currency, and chain-of-custody belongs in the case packet. The same is true for cleaning validation excerpts that support PDE/MACO statements when quality questions arise. Make these artifacts discoverable in the TMF and reference them in the PSMF so inspectors see one coherent system rather than scattered documents.

Systems and Validation: How to Prove You Control Your Data

Regulators increasingly focus on whether your systems work, not merely whether SOPs exist. Your safety database and analytics stack must be validated to a fit-for-purpose level under Part 11/Annex 11. That means defined user requirements, risk-based testing, traceability matrices, role-based access, and audit trails that actually get reviewed. Time synchronization matters—if your alarm server and database are 10 minutes apart, your clock-start calculations will drift. For analytics, version-lock code (Git), containerize, and archive data cuts with checksums; re-runs should reproduce the same hashes. ALCOA principles should be obvious in your artifacts: who performed which coding change, when; who merged potential duplicates; and which version of MedDRA and E2B dictionary was in force.

On the “edges,” show how PV integrates with manufacturing/quality. Many safety questions begin with “could this be a lot problem?” Maintain lot-to-site mapping, cold chain logs, and concise quality memos with representative PDE/MACO examples. When laboratory criteria define a case (e.g., assays for anti-PF4 or troponin), attach method sheets and LOD/LOQ so inclusion/exclusion is transparent. Finally, tie all of this to governance: a weekly signal meeting that reviews PRR/ROR/EBGM screens, O/E tallies, and any SCCS or cohort updates—and records decisions with owners and deadlines. This is the “living” proof that your framework is operational, not theoretical.

Signal Management to Label Change: A Step-by-Step, Inspection-Ready Path

Signals are hypotheses that require disciplined testing and documentation. Pre-declare your screens (e.g., PRR ≥2 with χ² ≥4 and n≥3; ROR 95% CI >1; EBGM lower bound >2) and your denominated follow-ups (O/E during biologically plausible windows, such as 0–7/8–21 days for myocarditis; 0–42 days for GBS). Confirm with SCCS or cohort designs; prespecify decision thresholds (e.g., SCCS IRR lower bound >1.5 in the primary window plus a clinically relevant absolute risk difference, ≥2 per 100,000 doses). Throughout, log quality context that could otherwise confuse causality—lots in shelf life, cold-chain TIR ≥99.5%, and representative PDE/MACO controls unchanged. If labs contribute to adjudication, include LOD/LOQ and calibration certificates. When a signal is confirmed, update the RMP, revise labeling and HCP guidance, and file an eCTD supplement that cites methods, outputs, and code hashes. Communication must use denominators and absolute risks to preserve trust.

Inspections and Readiness: What Inspectors Ask—and How to Answer

Inspectors want to follow a straight line from data to decision. Prepare a “read-me-first” index that maps SOPs → intake/coding rules → database cuts (date, software versions) → analytics code (commit IDs/container hashes) → outputs (screen logs, O/E worksheets, SCCS tables) → decision minutes → label/RMP changes. Demonstrate that your system is monitored, not just documented: monthly audit-trail reviews of privileged actions (case merges, threshold changes); KPI dashboards for timeliness (% valid ICSRs triaged in 24 hours), completeness (ICSR data-element score), and reproducibility (hash matches on re-runs). Show that you train to the system with role-based curricula and drills—e.g., simulated data-cut to filing within 5 business days—and that gaps become CAPAs with effectiveness checks. Keep quality appendices ready: representative PDE 3 mg/day; MACO 1.0–1.2 µg/25 cm²; method sheets with LOD / LOQ when assays drive inclusion. If asked “why did you not signal earlier?”, your answer should point to pre-declared thresholds, MaxSPRT boundary plots (if using rapid cycle analysis), and minutes demonstrating timely review.

Case Study (Hypothetical): Label Update Completed in Six Weeks Without Findings

Context. A sponsor detects a myocarditis pattern in males 12–29 within 7 days of dose 2. Screen. PRR 3.1 (χ² 9.8), EB05 2.4 across two spontaneous-report sources. O/E. 1.2 M doses administered; background 2.1/100,000 person-years → expected 0.48 in 7 days; observed 6 adjudicated Brighton Level 1–2 cases → O/E 12.5. Confirm. SCCS IRR 4.6 (95% CI 2.9–7.1) for Days 0–7; IRR 1.8 (1.1–3.0) for Days 8–21; absolute excess ≈ 3.4 per 100,000 second doses in young males. Action. RMP updated (important identified risk), label revised, Dear HCP communication issued with denominators. Quality context. Lots within shelf life; cold-chain TIR 99.6%; representative PDE/MACO unchanged; troponin method sheet attached (assay LOD 1.2 ng/L; LOQ 3.8 ng/L). Inspection. An unannounced GVP inspection finds no critical findings; the inspector notes strong traceability from raw data to decision.

Putting It All Together

The framework is manageable when you turn guidance into living controls. Map your obligations, validate your systems, pre-declare thresholds, practice the handoffs, and keep quality context at your fingertips. If your PSMF tells a coherent story and your TMF proves it with ALCOA discipline—plus transparent LOD/LOQ where labs matter and representative PDE/MACO where hygiene is questioned—you will make timely, defensible decisions and withstand inspection.

Pharmacovigilance for COVID-19 and Future Vaccines

Build the Right Pharmacovigilance Architecture: From Intake to Evidence You Can Defend

Post-marketing pharmacovigilance (PV) for COVID-19 vaccines—and for whatever comes next—requires a layered system that converts raw reports into defensible evidence. Start with intake and case processing that can scale: Individual Case Safety Reports (ICSRs) arrive via portals, email, call centers, and partner regulators. Your safety database should enforce E2B(R3) structure, MedDRA version control, and role-based access. Minimum case validity (identifiable patient, reporter, suspect product, and event) must be checked within 24 hours for seriousness triage. De-duplication rules (e.g., match on age/sex/onset/lot) are essential when media attention drives duplicate submissions. All edits and code changes must carry time-stamped audit trails consistent with Part 11/Annex 11, with ALCOA discipline visible in exported PDFs and XML acknowledgments filed to the TMF.

Once intake is stable, stitch passive reports to active, denominated datasets (claims/EHR, immunization registries) via privacy-preserving linkage. This lets you move from “someone noticed” to “how often relative to background.” Set up a governance cadence that blends clinical, epidemiology, statistics, quality, and regulatory. Every candidate signal should have a reproducible path: disproportionality screen → observed-versus-expected (O/E) check → sequential monitoring if needed → confirmatory study design (e.g., SCCS). Keep a one-page system map in your PV System Master File (PSMF) that links SOPs, databases, code repositories, and decision logs. For practical, regulator-aligned templates that speed SOP drafting, many teams adapt examples from PharmaSOP.in. For high-level public expectations and terminology you should mirror, consult the U.S. FDA.

COVID-19–Specific Practices That Should Become Standard: Speed, Adjudication, and Transparent Numbers

COVID-19 compressed safety decision cycles from months to days. Three practices deserve to persist. First, rapid cycle analysis (RCA) that updates weekly allowed earlier detection of real imbalances while controlling false positives; your protocol should pre-declare cadence, risk windows (e.g., myocarditis 0–7 and 8–21 days), and alpha-spending rules. Second, adjudication panels using Brighton Collaboration definitions turned noisy narratives into graded diagnostic certainty; maintain specialty panels (e.g., cardiology/neurology/hematology) and train them on uniform checklists. Third, transparent numbers build trust: when case definitions depend on biomarkers, state analytical capability—e.g., high-sensitivity troponin I LOD 1.2 ng/L and LOQ 3.8 ng/L for myocarditis confirmation; D-dimer assay LOD/LOQ for thrombotic events if relevant.

Quality context also matters. Reviewers routinely ask if manufacturing or hygiene could confound a safety pattern. Keep a succinct appendix that cites representative PDE (e.g., 3 mg/day for a residual solvent) and cleaning validation MACO limits (e.g., 1.0–1.2 µg/25 cm²) for the products and sites involved. Even though these are not “safety signals,” they reassure assessors that non-biological explanations (e.g., contamination) are unlikely, letting the analysis focus on biology and epidemiology rather than speculation.

Data Integrity, Dashboards, and What to Trend Every Month

A PV system that cannot show its own health will struggle in inspection. Define data-quality checks at intake (missing seriousness, impossible onset dates), coding (MedDRA drift), and analytics (version-locked code, reproducible seeds). Trend KPIs monthly and present them at Safety Governance: case validity within 24 hours, follow-up rate at 14 days, de-duplication yield, PRR screens reviewed on schedule, RCA boundary crossings, and time-to-decision for label actions. Implement a “completeness score” for ICSRs and route outliers to retraining. Keep external context visible by tagging media spikes and policy changes so you can explain bursts of reports without over-reacting.

Finally, make traceability obvious. Archive database cuts with date/time, software versions, and checksums; store adjudication minutes and decision memos in the TMF with cross-links to datasets and code. Run quarterly audit-trail reviews for privileged actions (case merges, code changes). When inspectors arrive, they should see a living system, not a static binder.

From Signal to Causality: PRR/ROR/EBGM → O/E → RCA → SCCS

Screening starts in spontaneous reports with disproportionality metrics. Pre-declare thresholds such as PRR ≥ 2 with χ² ≥ 4 and n ≥ 3; ROR with 95% CI excluding 1; and EBGM with lower bound (e.g., EB05) >2. These are hypothesis generators, not verdicts. Next, check observed versus expected using stratified background rates. Example (dummy): in one week, 1,200,000 second doses are administered to males 12–29; background myocarditis is 2.1/100,000 person-years. Expected in a 7-day window ≈ 1,200,000 × (7/365) × (2.1/100,000) ≈ 0.48. If six adjudicated Level 1–2 cases occur, O/E ≈ 12.5—strongly suggestive. If the program requires near-real-time oversight, initiate rapid cycle analysis (RCA) with MaxSPRT boundaries that control type I error across weekly looks. Confirm with self-controlled case series (SCCS), which compares incidence during risk windows (e.g., 0–7, 8–21 days) with control time within the same person, inherently controlling for fixed confounders. Declare how results drive actions: label updates, Risk Management Plan amendments, targeted studies, or enhanced monitoring.

Where laboratory markers define a case, keep the analytics transparent: assay LOD/LOQ, calibration certificates, and chain-of-custody for any central retesting. Maintain batch/lot traceability linking cases to distribution records; when regulators ask whether handling or hygiene could explain patterns, show that lots were in shelf life and under state-of-control with representative PDE and MACO examples already documented.

Case Study (Hypothetical): A Six-Week Path From Rumor to Label Action

Week 1–2: Passive screen. A cluster of myocarditis reports emerges in males 12–29, typically 2–4 days after dose 2; PRR 3.1 (χ² 9.8) and EB05 2.4. Narratives show chest pain and elevated high-sensitivity troponin I (above LOQ 3.8 ng/L). Week 3: O/E. 1.2 M second doses administered to males 12–29; expected 0.48 cases in 7 days; observed 6 adjudicated Level 1–2 → O/E 12.5. Week 4–5: RCA boundary crossed. MaxSPRT flags Days 0–7; clinical adjudication panel confirms Brighton levels. Week 6: SCCS. IRR 4.6 (2.9–7.1) for Days 0–7; IRR 1.8 (1.1–3.0) for Days 8–21. Action: label and RMP updated; Dear HCP communication drafted with absolute risks (“~12 per million second doses in young males within 7 days”) and guidance. Quality cross-check: lots in specification; cold-chain logs in range; representative PDE 3 mg/day and MACO 1.0–1.2 µg/25 cm² unchanged; no non-biological confounders found.

Future-Proofing: Governance for Next-Gen Platforms and Pandemics

mRNA, protein-adjuvant, and vector platforms will evolve; your PV governance should be ready before the next emergency. Pre-register AESIs by platform (e.g., myocarditis for mRNA, TTS for adenovirus vectors), their risk windows, and diagnostic packages. Maintain standing adjudication panels and reserve contracts for data access (claims/EHR/registries) with pre-approved protocols, so RCA and SCCS can start on Day 1. Keep communication templates that explain signal logic in plain language, include denominators, and link to public resources. Codify how manufacturing and distribution context is checked for every signal so quality questions do not derail medical decision-making.

Most importantly, make the record easy to follow. In your TMF and PSMF, keep a crosswalk that shows SOPs → data cuts → code → outputs → decisions → labeling. Version-lock code, archive database snapshots with checksums, and run scheduled audit-trail reviews. For method calibration, run periodic “negative control” screens to ensure the system is not over-signaling. When a real signal emerges, the combination of transparent thresholds, rapid analytics, clean documentation, and clear quality context will let you act quickly without sacrificing rigor.